Article: Modelling and diagnosis method of power transformer sound characteristics by integrating sparse representation and deep autoencoder network Journal: International Journal of Innovation and Sustainable Development (IJISD) 2026 Vol.20 No.7 pp.18 - 35 Abstract: This paper combines sparse representation with deep autoencoder (DAE) to propose a method for power transformer sound modelling and fault diagnosis. When a power transformer malfunctions, it is usually accompanied by abnormal sounds. During the operation of transformers, the sound signals generated by structural vibrations such as iron cores and windings contain rich equipment status information. Accurate detection and analysis of sound signals can achieve mechanical fault diagnosis of transformers. Traditional detection methods are no longer able to meet the constantly changing needs of the electrical power system (EPS). This paper constructs a fault sound recognition model that combines sparse representation and DAE. This model optimises the encoding process by introducing sparse constraints, thereby enhancing the ability to extract key fault signal features. The results indicate that the model has advantages in effective feature extraction, diagnostic speed, and fault diagnosis accuracy. Inderscience Publishers - linking academia, business and industry through research

Title: Modelling and diagnosis method of power transformer sound characteristics by integrating sparse representation and deep autoencoder network

Authors: Dong Li; Kuo Liu; Congcong Guo; Qingqing Gong

Addresses: Nanyang Power Supply of State Grid Henan Electric Power Company, Nanyang, 473000, China ' Nanyang Power Supply of State Grid Henan Electric Power Company, Nanyang, 473000, China ' Nanyang Power Supply of State Grid Henan Electric Power Company, Nanyang, 473000, China ' Nanyang Power Supply of State Grid Henan Electric Power Company, Nanyang, 473000, China

Abstract: This paper combines sparse representation with deep autoencoder (DAE) to propose a method for power transformer sound modelling and fault diagnosis. When a power transformer malfunctions, it is usually accompanied by abnormal sounds. During the operation of transformers, the sound signals generated by structural vibrations such as iron cores and windings contain rich equipment status information. Accurate detection and analysis of sound signals can achieve mechanical fault diagnosis of transformers. Traditional detection methods are no longer able to meet the constantly changing needs of the electrical power system (EPS). This paper constructs a fault sound recognition model that combines sparse representation and DAE. This model optimises the encoding process by introducing sparse constraints, thereby enhancing the ability to extract key fault signal features. The results indicate that the model has advantages in effective feature extraction, diagnostic speed, and fault diagnosis accuracy.

Keywords: deep learning; sparse representation; DAE; deep autoencoder network; power transformer; sound characteristics; fault diagnosis.

DOI: 10.1504/IJISD.2026.152281

International Journal of Innovation and Sustainable Development, 2026 Vol.20 No.7, pp.18 - 35

Received: 28 Oct 2025
Accepted: 23 Jan 2026
Published online: 13 Mar 2026 *

Title: Modelling and diagnosis method of power transformer sound characteristics by integrating sparse representation and deep autoencoder network

Keep up-to-date