Power Module Junction Temperature Prediction Using NTC Network and Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting the junction temperature of power semiconductor modules are limited by the need for invasive procedures, low sampling rates, difficulty in measuring electric current through parallel chips, complex and insensitive measurement circuits, and temperature sensors that do not directly contact the chip, making real-time and accurate monitoring challenging.
Innovation Solution
A method using a negative temperature coefficient (NTC) thermistor network and neural network model to predict junction temperature by monitoring temperature across the power module, training the model with data from finite element simulations or experiments, and deploying it on a controller for real-time online detection and aging evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the optical method (infrared thermal imager) is used to detect junction temperature, then two-dimensional temperature distribution of multiple chips can be obtained, but the sampling rate is limited and cannot meet real-time measurement needs
Solution Approach 1:
The patent replaces the optical/infrared detection system with an electrical parameter-based detection system. By measuring electrical parameters (such as threshold voltage, turn-off delay time, or on-resistance) that change with temperature, the system achieves fast response and real-time monitoring without the sampling rate limitations of infrared thermal imagers.
Solution Approach 2:
The patent introduces thermosensitive electrical parameters as intermediaries to indirectly measure junction temperature. These electrical parameters serve as mediators that correlate with temperature, enabling real-time detection without direct thermal contact or optical measurement limitations.
2Measurement precision
If the thermal network model method is used to calculate junction temperature, then junction temperature can be obtained according to substrate temperature and thermal resistance model, but accurate loss online identification is difficult due to inability to measure electric current through parallel chips
Solution Approach 1:
The patent enables the power module to self-diagnose its thermal state by measuring electrical parameters that inherently change with temperature. The module uses its own electrical characteristics as sensors, eliminating the need for external current measurement equipment and complex thermal models.
Solution Approach 2:
The patent exploits the natural changes in electrical parameters (threshold voltage, turn-off delay time, on-resistance) with temperature. By monitoring these parameter changes, the system directly infers junction temperature without needing to measure current through parallel chips or maintain complex thermal resistance models.
3Productivity
If the thermosensitive electric parameter method is used to obtain junction temperature, then fast response and easy on-line detection are achieved, but measurement circuits are complex and sensitivity is relatively low
Solution Approach 1:
The patent makes existing electrical measurement circuits serve dual purposes: both for normal power module operation control and for junction temperature detection. The same circuits that control power switching also measure thermosensitive parameters, eliminating the need for separate complex measurement circuits.
4Ease of operation
If the physical contact method (temperature sensors) is used to measure temperature, then temperature information of contacting parts can be obtained, but the temperature sensor cannot directly contact the chip so the measured temperature differs from actual junction temperature
Solution Approach 1:
The patent replaces physical thermal contact measurement with electrical parameter measurement. By measuring electrical properties that change with temperature, the system obtains accurate junction temperature information without physical contact, thus achieving both ease of operation and measurement precision.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Accurately predicts real-time junction temperature with high precision, identifies potential failures, and evaluates the State Of Health (SOH) of the module, improving prediction accuracy and enabling early warning of aging phenomena.
Implementation Method 1
A method using a negative temperature coefficient (NTC) thermistor network and neural network model to predict junction temperature by monitoring temperature across the power module
Data Source
AI summary
The present disclosure belongs to the technical field of power electronic converters, and discloses a method and a system for predicting a junction temperature of a power semiconductor module in the full life cycle and a terminal. The method includes the steps: arranging an NTC thermistor network to monitor the temperature of each area inside the power module when the power module works; obtaining data for training the neural network by utilizing finite element simulation or experiments, and building a neural network model among the temperature of the NTC resistor network, a water flow rate, an aging factor and the junction temperature of the chip under working conditions. The present disclosure improves the junction temperature prediction accuracy of areas with relatively large errors comprehensively and realizes the high-precision junction temperature prediction under all working conditions.