ARTIFICAL INTELLIGENCE Predictive maintenance of power electronic devices

Related Vendors

Diotec Semiconductor AG

Rogers Corporation

DOWA HD Europe GmbH

What if you could estimate the real age of your power converter and know when to replace it? Predictive maintenance is an approach that uses artificial intelligence and machine learning algorithms to estimate maintenance schedules for power electronic devices and systems. The article explains predictive maintenance in power electronics.

In mainstream media, the use of AI and ML is limited to computer networking and the IT industry. The semiconductor manufacturing industry extensively implements such technologies to automate chip production. In power electronics, AI may seem to be unimportant. However, engineers have started to use AI and ML in advanced power electronic systems to predict long-term results and test device functionalities in various scenarios.

An overview of predictive maintenance

Predictive maintenance is a technique that utilizes sensors to monitor the real-time working conditions of equipment in service, estimating the optimal maintenance schedule. In simple words, predictive maintenance is a condition-based monitoring system that uses sensors to estimate when maintenance should be performed. It uses sensors to collect data from hardware and then processes such data using advanced machine learning and AI algorithms to predict potential failure states.

In comparison to preventive maintenance, predictive maintenance estimates an optimal maintenance schedule rather than performing maintenance at predetermined intervals. Similarly, predictive maintenance does not follow the principles of corrective maintenance, which performs maintenance at the time of failure. The power electronics industry tends to perform preventive and corrective maintenance. These approaches lead to system failure, unplanned downtime, and increased operational costs.

Planned maintenance schedules do not hinder system operation much. Secondary or replacement-type equipment can be deployed to run the system. The approach emphasizes a device’s continuous operation without failure or downtime to reduce operational costs. Simulation and predictive maintenance techniques are extensively implemented in the renewable energy domain including solar power, EVs, wind energy, and hydropower.

Predictive maintenance in power electronic devices

Power electronics involve high-current and voltage applications. In such cases, devices are prone to excessive heat dissipation. Such scenarios can sometimes cause equipment failure. As a result, predictive maintenance in power electronic systems and devices continuously monitors parameters like voltage, current, frequency, temperature, and many other parameters.

In power electronics, predictive maintenance can be applied to power converters, rectifiers, inverters, drives, uninterrupted power supplies, and many other components. As mentioned above, the use of predictive maintenance is common in the renewable industry.

POWER ELECTRONICS

Inverter vs converter: What's the difference?

How does predictive maintenance work?

The predictive maintenance procedure is executed within three main stages.

Sensor data acquisition: Predictive maintenance uses complex data generated by devices and IoT sensors. Data can also be virtually generated by software.

Data transformation: Complex data is transformed for analysis. Simply put, data transformation can be the removal of noise, approximation of values, complex computations, and elimination of irrelevant values.

AI and Machine Learning: AI and ML models train and validate data to refine data. The generated data is compared with the historical set. AI and ML algorithms perform further analysis to identify patterns and predict failure trends and maintenance schedules.

Features of predictive maintenance for power electronics

Predictive maintenance processes ensure connectivity between sensors, IoT devices, edge devices, and cloud platforms.

IoT integration: Integration of IoT devices enables remote monitoring, management, and protection of power electronic systems and their data.

Cloud-based analytics: Cloud-based platforms ensure that data is analyzed in real-time and maintenance schedules are deployed in the system.

Continuous updates: Existing AI and ML models are continuously updated with new data to adapt to changing operating and environmental conditions, further reducing the risk of downtime.

Benefits of predictive maintenance for power electronics

Predictive maintenance enables enterprises and industries to make planned data-driven decisions. It improves the efficiency, reliability, and performance of power electronic devices and systems.

Enhanced safety

Corrective maintenance after failure can risk lives due to the high-current and voltage nature of power electronic environments. As a result, predictive maintenance prevents such “hazardous repair situations” and accidents.

Reduced failure

With the help of predictive maintenance, engineers can replace components, adjust load, change operating conditions, and implement advanced cooling mechanisms to eliminate chances of failure and downtime.

Cost-cutting

Predictive maintenance optimizes the frequency of scheduled maintenance. Unnecessary maintenance procedures at regular predetermined intervals are no longer necessary. High-maintenance tasks are prioritized instead.

AI techniques to implement predictive maintenance in power electronics

In traditional processes, maintenance used to be performed on systems based on ratings, past sets of data, and manual needs. The section lists a few popular AI and ML techniques used in predictive maintenance for power electronics.

Digital twin

DT is a virtual version of a real-time physical object or intended device, place, system, process, or person. The entire predictive maintenance set-up can be done on software using the simulation of power devices and virtual sensors for various scenarios.

Real-time power electronic hardware can also be used with DT software to perform predictive maintenance. Such cases are examples of hardware-in-the-loop simulation, where real hardware is tested for virtual scenarios. When DT software receives data about power electronic devices from sensors, it estimates maintenance cycles through advanced AI or ML algorithms.

DIGITAL TWINS

Use of digital twins in power electronics development

Anomaly detection

When the behavior of a device or a system deviates from traditional behavior, anomalies occur. Anomaly detection is an approach that relies on such principles. Statistical methods, neural networks, and clustering algorithms are key anomaly detection techniques. There are two modes of operation: normal mode and abnormal mode.

For example, in power electronics, anomalies can occur when a device operates at very different voltage and current levels than normal or excessive voltage fluctuations occur during the operation. AI and ML algorithms automatically detect anomalies. However, such a method is not reliable for root-cause analysis.

Feature engineering

Organizations use feature engineering for predictive maintenance of power electronics devices and systems. Feature engineering starts with relevant data collection through deployed sensors, IoT devices, databases, and maintenance logs. Some software is capable of automatically generating daily data for predictive maintenance.

Data is processed and transformed for modeling. In power electronics, features could be voltage, current, power, frequency, temperature, and many more characteristics. Meaningful features are extracted from the data. AI and ML algorithms select features to make predictions about power electronic devices and systems.

Causal AI

Causal AI technology is becoming increasingly popular in the tech industry for its accuracy compared to vague responses by generative AI. Causal AI relies on a cause-to-effect relationship between two entities and root-cause analysis through power electronics expert knowledge and data-driven insights.

For example, a converter has suddenly reduced its efficiency. Generative AI collects data on the internet to form responses. On the other hand, causal AI uses ML models to determine the reason and point out the components for maintenance. As of 2024, only a few software tools enable causal AI-based predictive maintenance for real-time power electronic systems.

References

• https://www.researchgate.net/publication/379959666_Real-Time_Predictive_Maintenance_of_Power_Electronics_Systems_using_Machine_Learning_and_IoT_Integration

• https://resources.pcb.cadence.com/blog/2023-predictive-maintenance-in-the-electronics-industry

(ID:50263276)