Enterprise AI Analysis
Module Parasitics-Based Current and Temperature Sensing Using Explainable Neural Networks
This paper examines the application of simple neural networks for current measurement and the determination of the junction temperature in power semiconductor modules. On the one hand, the focus was not on the use of conventional sensors such as current sensors or temperature sensors, but rather on utilising parasitic components within the power semiconductor module, from which useful signals can be extracted. Namely, these are the voltage across parasitic inductances in a module, the semiconductor's on-state voltage, and its turn-on delay time. Because these signals are often affected by other parameters, the desired information must be extracted, which was found to be an application case for artificial neural networks. On the other hand, the application of ANNs in the simplest and most effective way possible was presented. Furthermore, a method is introduced that takes a first step towards the interpretability of neural networks in a straightforward manner to overcome the main drawback for the user-the usual black-box structure of neural networks.
Executive Impact Summary
This research demonstrates how AI-enhanced sensorless monitoring leverages inherent module parasitics for precise current and temperature sensing, offering a pathway to reduced costs, improved reliability, and predictive maintenance. Critical insights include:
±0%
Current Sensing Accuracy
0 K
Junction Temp Accuracy
0
Indirect Signals Used
By intelligently processing parasitic signals, neural networks can overcome measurement challenges, providing accurate data crucial for advanced diagnostics and operational optimization. The proposed explainability method allows validation of AI models, ensuring they are physics-grounded rather than opaque black-boxes.
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
AI-Powered Parasitic Current Sensing
±1.5% Current Sensing Error ReductionNeural networks effectively compensate bond wire temperature influence, reducing current sensing error to ±1.5% using parasitic signals derived from module inductances and resistances, enhancing accuracy and robustness.
Enterprise Process Flow: Explainable AI Methodology
Select Representative Data Sample
→
Maintain All But One Input Constant
→
Vary Single Input Parameter Systematically
→
Observe and Plot Network Output
→
Interpret Resulting Characteristic Curves
→
Assess Physical Consistency and Generalization
To understand if an ANN has truly learned underlying physics, a sensitivity test systematically varies one input while holding others constant, plotting the output. This reveals characteristic curves, allowing assessment of physical consistency and generalization capabilities beyond simple memorization.
| Training Data Quality | Current Sensing 95th Percentile Error | ANN Behavior & Interpretability |
|---|---|---|
| High Quality (Smooth Input Data) | 9.41 A |
|
| Poor Quality (Noisy Input Data) | 28.5 A |
|
The quality of training data is paramount for ANN performance and interpretability. High-quality, smooth data enables ANNs to learn robust physical correlations, leading to lower errors and explainable behavior. Conversely, noisy data results in significantly higher errors, inconsistent learning, and black-box-like behavior, undermining trust and practical utility. |
||
Critical Dependency: Temperature Sensing without td,on
Problem: Relying solely on on-state drain-source voltage (VDS,on) for junction temperature (TJ) estimation, especially without the crucial turn-on delay (td,on) signal, proves highly problematic. VDS,on is sensitive to both current and temperature, and its typical measurement accuracy (e.g., 1.7% full scale) is insufficient when used in isolation.
Solution: Without td,on, the ANN struggles to extract reliable temperature information from VDS,on. Simulated scenarios with realistic VDS,on inaccuracies lead to a 95th percentile error of 44.1 K and a mean error of 27.0 K for TJ. This makes temperature estimation fundamentally unreliable and impractical for condition monitoring.
Impact: This case highlights that simply providing more input signals to an ANN does not guarantee robust learning if those signals lack sufficient quality or intrinsic temperature sensitivity (like td,on). A high-quality, dedicated temperature-sensitive signal is often essential for accurate junction temperature estimation, even with AI processing.
Advanced ROI Calculator
Estimate your potential savings and efficiency gains by implementing AI-enhanced sensorless monitoring in your operations.
Annual Cost Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
Our phased approach ensures a smooth integration of AI-driven sensorless monitoring into your existing infrastructure, maximizing impact and minimizing disruption.
Phase 1: Discovery & Strategy
In-depth analysis of your current power module monitoring systems, identification of key performance indicators, and development of a tailored AI strategy utilizing parasitic signals.
Phase 2: Data Acquisition & Preprocessing
Establishment of efficient parasitic signal capture (VLP, VDS,on, td,on) and implementation of robust data cleaning and smoothing pipelines for optimal ANN training.
Phase 3: Model Development & Training
Design and training of custom neural networks using your specific operational data, focusing on achieving high accuracy for current and junction temperature estimation, validated with explainability tests.
Phase 4: Integration & Deployment
Seamless integration of the AI monitoring solution into your enterprise systems, including real-time data feeds and dashboard visualization, with ongoing calibration and refinement.
Phase 5: Performance Monitoring & Optimization
Continuous monitoring of the AI system's performance, proactive adjustments, and iterative improvements to ensure sustained accuracy and maximum operational benefits.
Ready to Transform Your Operations?
Unlock predictive maintenance, optimize performance, and drive significant cost savings with our bespoke AI solutions.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat