Ultra-Low Power Neuromorphic Front-End Chip for Predictive Maintenance
VibroSense, a Tiny AI application-specific chip based on the Neuromorphic Analog Signal Processor, performs pre-processing of vibration data on the sensor node and addresses the major challenge of Industrial IoT: amount of data to be transmitted, stored and processed.
Sensor Raw Data Pre-Processing with Neural Networks Saves A Fortune!
Predictive Maintenance can be used to achieve the best possible upkeep of industrial equipment. It involves gathering and analyzing data from sensors such as vibration and acoustic. The machines being monitored are equipped with vibration sensors that collect data and transmit it to a local server or cloud for storage and processing. The vast amount of data generated by vibration sensors during operations makes Predictive Maintenance costly due to the expenses associated with data transmission, storage, and processing.
VibroSense chip solves the problem by reducing the amount of data transmitted and making Predictive Maintenance more accessible and cost-effective on a global scale.
1Always-on vibration sensors produce huge amounts of data
2IIoT systems require low bandwidth communications
3Predictive Maintenance solutions are resource-hungry and expensive
1Raw data pre-processing on-sensor reduces data flow by 1000 times
2NASP transmitting only small data patterns supports low bandwidth long distance communications
3Processing significantly less data reduces OPEX and Cloud TCO
Vibration sensors are typically attached to equipment to measure the vibrations that rotating parts generate. These sensors usually work at frequency bandwidth of up to 20KHz to ensure accurate prediction of mechanical failures, which is critical for Predictive Maintenance applications. However, the high-frequency signals create large amounts of data that must be sampled and transmitted to the cloud for further processing, generating alerts, and providing maintenance recommendations.
VibroSense replaces high-volume vibration data with small patterns that are transmitted to the cloud instead. The patterns, uniquely corresponding to specific signals, contain information from the vibrations and are 1000 times smaller in size, effectively reducing the amount of data that needs to be transmitted. Despite the reduced size, the information contained in these patterns is still sufficient for reliable Predictive Maintenance.
Vibration-based condition monitoring is a fundamental Predictive Maintenance technique that is used to detect machine failures. By analyzing vibrations, it is possible to identify a range of machinery problems such as shaft unbalance and misalignment, bearing failures, gear wear, cracks, looseness and more. With VibroSense, the vibration signals are pre-processed on the sensor node, which provides significant global savings and solves the bandwidth and power challenges that are commonly faced by the IIoT industry.
Power consumption is a critical factor in battery-powered Industrial IoT applications, with data transmission accounting for 85-99% of the total power consumption in wireless sensors. POLYN’s VibroSense ultra-low power AI chip uses NASP, the patented technology for pattern extraction from large-size vibrations, reducing the data flow by 1000 times. The operations of the always-on chip consume only 100 µW of power, resulting in significant energy savings. This offers several major benefits: improvements in sensor logistics, support of energy harvesting devices and LPWA (low power wide area) data transmission techniques.