Brushless (BL) DC motors, or BLDC motors, have the great advantage of not having brushes and commutators, which are the biggest disadvantages of conventional DC motors, so they have less wear and tear, less energy loss, and higher maintenance and energy efficiency. Accordingly, BLDC motors are mainly adopted in the development of most products and facilities, such as drones, robots, mobility, industrial automation, home appliances, and mechanical equipment.

Although these BLDC motors are highly reliable, they are not free from failures and damages such as bearing failure, overheating, electrical problems, and mechanical damage. In particular, in products such as drones and mobility, motor damage can lead to serious accidents, so functions for failure prediction and predictive maintenance are becoming increasingly important.

In the past, predictive maintenance of motors and equipment was installed outside the product, and sensing, data collection, fault diagnosis, and notification were performed in separate processes. This required time and money to select a company that provides predictive maintenance solutions and products, and to introduce products. In addition, inefficiencies arose during the development process due to increased system complexity.

Recent advances in AI technology and improvements in semiconductor performance are showing signs of integrating these fault diagnosis mechanisms at the edge. MCU makers such as ST and TI are enhancing the convenience of motor control by implementing solutions such as fault diagnosis and monitoring within the ‘edge AI’ function.

■ ST, Support for Motor Diagnosis Edge AI Function
Late last year, ST introduced the STM32N6 MCU with an NPU capable of up to 600 GOPs. We are introducing the first AI MCU based on the 800MHz Arm Cortex-M55 that aims to implement high-performance edge AI performance in MCU.

ST’s Manager Moon Hyun-soo introduced edge AI use cases that utilize STM32N6 MCUs as well as software tools such as STM32Cube.AI and NanoEdge AI Studio, saying, “A representative example is predictive maintenance implemented by additionally installing modules on the motor.” He added, “The operating principle is that the vibration of the motor is detected with a MEMS sensor in this module, and even minute abnormal vibrations that cannot be detected by humans are detected.”

He introduced a system that notifies maintenance time before breakdown using vibration data from these microscopic motors, and added that aftermarket edge AI devices that simply add edge AI modules to existing devices are characterized by low investment costs and very low power consumption.

ST is also offering, through its Partner Design Services, Tessolve’s Predictive Maintenance (PM) AI-based demo kit that uses an accelerometer to identify motor anomalies in machines using the NanoEdge AI library.

The Tessolve PM demo is a use case that supports motor anomaly detection and real-time anomaly transmission based on LoRa communication on low-power MCUs and NPUs that enable AI models to run.

■ TI, Optimizing System Fault Detection Function
TI also launched an MCU product with integrated NPU around the same time as ST, leading the AI MCU competition.He stood in opposition to ST in the island.

The TMS320F28P55x series C2000™ MCUs are the first real-time MCU portfolio to integrate a neural processing unit (NPU), delivering high-accuracy and low-latency fault detection. They are based on the Arm Cortex-M7 with NPU performance of 600 to 1200 MOPS.

Application examples of these TI AI MCUs include arc fault detection in solar and energy storage systems, as well as motor bearing fault detection for predictive maintenance.

TI emphasized that the TMS320F28P55x series enables the system to detect errors with greater than 99% accuracy through NPU, predicting failures in equipment and working environments, and explained that the CNN model for motor bearing and arc fault detection can learn complex patterns from raw sensor data such as vibration signals, and then detect subtle changes that indicate bearing faults.

CNN models excel in sensor data analysis for fault detection and predictive maintenance because they can learn autonomously from raw or preprocessed sensor data, such as motor vibration signals, solar DC current, or battery voltage and current.

In fact, a study titled 'Research on Fault Diagnosis Algorithm for Thermal Damage to BLDC Motors' published in the Journal of the Korean Society of Mechanical Engineers in March 2025 also found that the CNN model showed the best performance in fault diagnosis of drone BLDC motors.