Metal particle defect detection in gas insulated switchgear (GIS) primarily relies on piezoelectric lead zirconate titanate (PZT) sensing systems, which struggle to meet power equipment’s high-level operation and maintenance requirements. In recent years, numerous research teams have investigated optical fiber interferometric acoustic emission sensing technology. However, there remains a lack of theoretical solutions and implementations for the optimization design of optical fiber sensing units, with experimental tests on actual GIS platforms being scarce. This paper presents an optical fiber acoustic emission sensing system based on Mach-Zehnder interference. It explores the frequency response characteristics of the optical fiber sensing unit by combining theoretical calculations with experimental tests and proposes an optimization design scheme for the optical fiber sensing unit targeting GIS metal particle detection. Furthermore, a metal particle defect detection experiment is conducted on a 126 kV GIS. The experimental results demonstrate that the Mach-Zehnder interference-based optical fiber acoustic emission sensing system exhibits superior detection capabilities to traditional PZT systems. Under identical excitation signals, the maximum response amplitude of the optical fiber is approximately 17.4 times that of a PZT system.