Investigation of artificial neural network algorithm based IGBT online condition monitoring. (September 2018)

• Record Type: Journal Article
• Authors: Sun, Xiaoman; Huang, Meng; Liu, Yi; Zha, Xiaoming
• Is Part Of: Microelectronics and reliability. Volume 88/90(2018)
• Page Start: 103
• Page End: 106
• Publication Date: 2018-09
• Subjects: IGBT reliability -- Lifetime -- IGBT online monitoring; Electronic apparatus and appliances -- Reliability -- Periodicals; Miniature electronic equipment -- Periodicals; Appareils électroniques -- Fiabilité -- Périodiques; Équipement électronique miniaturisé -- Périodiques; Electronic apparatus and appliances -- Reliability; Miniature electronic equipment; Periodicals; 621.3815
• DOI: 10.1016/j.microrel.2018.07.068
• Languages: English
• ISSNs: 0026-2714

Abstract

Abstract: Reliability of Insulated Gate Bipolar Transistor (IGBT) has drawn much attention in recent years. Online monitoring of IGBT is an effective mehod to improve IGBT operation reliability. State-of-the-art online monitoring methods for IGBT are based on thermal sensitive electrical parameters (TSEPs) extraction, but the TSEPs can be hardly obtained with required accuracy in practical application. This paper investigates Artificial Neural Network (ANN) based IGBT online monitoring method. DC link voltage and H-bridge output voltage, which are practical measurable parameters, are selected as the input of ANN. Both single input single output (SISO) and multiple input single output (MISO) neural networks are analysed and discussed. With the proposed method, the relationship of the practical measurable parameters and investigated TSEP, on-resistance of IGBT, can be established. By applying the existing criterion of TSEPs for the IGBT reliability, the prediction of the IGBT failure can be achieved. Simulations verify that the errors brought by the established model are within precision requirements. Highlights: Both single input single output (SISO) and multiple input single output (MISO) neural networks can be applied in the prediction for IGBT on resistance value within the accuracy requirement. Multiple input single output (MISO) neural networks is more accurate than single input single output (SISO) neural networks in the prediction.