Neuromorphic AI for always-on industrial monitoring

Neuromorphic AI uses spiking neural networks to process data in an event-driven way. Although this brain-inspired type of architecture has been widely researched, 42T has now shown how it can be applied in a practical, low-cost system for continuously monitoring motors, pumps, fans and other industrial equipment.

The new approach combines Innatera’s low-cost, ultra-low-power neuromorphic processor with sensor technology to detect emerging faults within a few seconds. It tackles one of the sector’s biggest challenges: detecting equipment faults early enough to prevent costly downtime or expensive system failures.

Paul Bearpark, head of electronics and software at 42T, said: “Continuous condition monitoring is already used to protect critical industrial equipment, but existing approaches are often expensive, power hungry, and complex to make it practical for monitoring every asset.

“Neuromorphic processing changes that equation by enabling a small, inexpensive and intelligent device to analyse sensor data locally and continuously. And for the first time, companies can extend predictive maintenance to assets that were simply too costly or complex to monitor before.”

The technology can be deployed either as a retrofit monitoring patch powered by a small battery or built into connected industrial devices. The monitoring module operates continuously while consuming 600 μW of power or less and has an expected system cost below US$10. This makes continuous low-power monitoring practical even in highly cost-sensitive applications.

42T has built a working demonstrator that uses a MEMS-based inertial measurement unit (IMU) to monitor vibration from an electric motor. The sensor data is processed locally by Innatera’s Pulsar neuromorphic microcontroller which has been trained to identify and classify specific changes in the motor’s behaviour, and to trigger an alert when an anomaly is detected.

The demonstrator integrates the IMU, Pulsar processor, alert function and power module, and has been showcased at major industry events including CES 2026 (Las Vegas, US), Embedded World 2026 (Nuremberg, Germany), and Edge AI London.

42T’s current demonstrator monitors motor vibration. However, the core approach can process a wide range of time-series data, including audio, current, voltage, power and temperature signals. It can also handle multiple sensor inputs to detect, for example, the vibration and audio changes caused by worn motor bearings.

Unlike conventional processors that continuously process streams of data, spiking neural processors use an event-driven architecture in which computation is triggered by changes in incoming signals. Innatera’s architecture, used in its Pulsar processor, is designed to deliver up to 100-fold faster processing and 500-fold lower power consumption compared with conventional AI processors.

The system can be trained to recognise specific types of faults, such as known changes in a machine’s vibration signature. Alternatively, it can learn the range of normal operation and flag any deviations from that baseline, potentially allowing previously unseen anomalies to be identified.

In addition to its industrial applications, 42T’s proposed approach could open up opportunities for always-on monitoring in consumer products. Potential applications include smart water-leak detection, predictive maintenance in household appliances such as washing machines and refrigerators, and health or fitness monitoring devices.

42t.com