Can AI Unlock the Future for Condition-Based Monitoring at the Edge?

In an era when equipment and energy costs are rising, industries are under increasing pressure to optimize their operations and reduce expenses. One effective approach is to minimize unplanned downtime and extend the lifespan of assets. Predictive maintenance strategies, such as condition-based monitoring (CbM), can streamline maintenance activities, reducing the need for emergency repairs and lowering overall maintenance costs. But can the addition of artificial intelligence (AI) enhance these efforts even further?

Wireless CbM can be implemented either through cloud-based or edge-based systems. While cloud-based solutions offer some advantages, system integrators must also consider factors like latency in decision-making, network usage, and power consumption/battery life. AI edge-based CbM addresses these concerns by processing data locally, at the source.

To successfully implement AI edge-based CbM in your factory, you need a system that enables you to:

1. Develop and deploy AI at the edge
2. Rapidly deploy wireless solutions to machines or test setups
3. Evaluate microelectronic mechanical system (MEMS) sensor technology for vibration monitoring
4. Assess communication technology for industrial wireless sensing
5. Accelerate asset monitoring and solution development
6. Evaluate and develop edge AI solutions for industrial CbM

To assist with this, a comprehensive evaluation solution was designed to revolutionize the approach to vibration monitoring, called Voyager4 . This MEMS-based wireless platform seamlessly integrates hardware, software, and mechanical designs, enabling customers to swiftly assess a complete signal chain when attached to machinery.

Voyager4 goes a step further by incorporating intelligence at the industrial edge, featuring a convolutional neural network (CNN) hardware accelerator. This advanced technology not only supports automatic machine anomaly detection but also enhances system-level benefits, such as extending sensor battery life. (See Figure 1)

Figure 1: Voyager4 Enclosure and PCB

Discovering Voyager4 

The ADG1634 CMOS switch, which is used to route the ADXL382 MEMS raw vibration data to either the MAX32666 BLE radio or the MAX78000 AI microcontroller.

The BLE microcontroller controls the single pole, double throw (SPDT) switch and can stream ADXL382 MEMS raw data to the host through either BLE or USB using the FTDI FT234XD-R.

The MAX17262 fuel gauge monitors battery current, and a nano power ADXL367 optional wake-up MEMS accelerometer is used to wake up the BLE radio from Deep Sleep mode in a high vibration shock event.

The MAX20335 PMIC features two buck regulators and three low-dropout regulators (LDOs). This PMIC allows the user to configure power domains corresponding to different Voyager4 operating modes. (See Figure 2)

Figure 2: Voyager4 Hardware System

Operating Principle 

1. ADXL382 raw vibration data is sent to the user over BLE or USB where this data is used to train an edge AI algorithm using MAX78000 tools.
2. AI model is synthesized into C code and the AI algorithm is sent via  USB update and stored in MAX78000 memory.
3. After this initial Voyager4 training phase:
   1. Edge AI algorithm will predict faulty or healthy machine operation based on the vibration data gathered.
   2. If faulty vibration data is predicted, a vibration anomaly alert is sent to the user over BLE, otherwise, the Voyager4 sensor does not use the radio and goes back to sleep. (See Figure 3)

Figure 3: Voyager4 Operating Principle

Battery Life

Harnessing edge AI can significantly extend battery life by up to 50% (Figure 4). When the evaluation kit operates in training mode, it consumes over 0.65 mW of power as the BLE remains active: advertising, connecting, and transmitting data once per hour.

In normal (AI) mode, the BLE radio's advertising, connection, and streaming features are disabled by default. Periodically, the MAX78000 wakes up to perform AI inference. If no anomaly is detected, Voyager4 returns to Deep Sleep mode.

Operating in AI mode, Voyager4's power consumption drops to approximately 0.3 mW, even with hourly sensor activity. Figure 5 illustrates that a sensor, which doesn't need to transmit raw BLE data, can reduce power consumption by up to 50%.

At a power consumption rate of 0.3 mW, the estimated battery life is an impressive seven years with two AA-sized 2.6 Ah LS14500 Saft batteries, or two years with a single 1500 mAh ASR00073 battery.

Figure 4: Voyager4 Average Power Consumption

Conclusion

The integration of AI at the edge in CbM represents a significant leap forward in industrial maintenance strategies. Voyager4, with its advanced MEMS-based wireless platform and CNN hardware accelerator, not only streamlines the deployment of wireless solutions but also optimizes power consumption, extending battery life and reducing operational costs. By enabling automatic anomaly detection and minimizing the need for constant data transmission, Voyager4 empowers industries to achieve more efficient and reliable asset health monitoring. This innovative platform is a key enabler for the factory of the future, where intelligent, edge-based systems play a crucial role in maintaining optimal performance and minimizing downtime. (See Figure 5).

Figure 5: Voyager4

For more information about predictive maintenance solutions, please visit:

• Landing Page:
  • Predictive Maintenance Solutions | Analog Devices 

• Evaluation Board:
  • Voyager4 Evaluation Board | Analog Devices
•  Technical Articles:
  • Factory of the Future: Designing Edge Sensors with Artificial Intelligence—Part 1 | Analog Devices
  • Factory of the Future: Making Decisions at the Edge Using Sensors with Artificial Intelligence—Part 2 | Analog Devices