In rapid development of industrial cyber-physical systems, safety and reliability are critical. Early fault detection which identifies safety and security issues at early stage is vital to ensure system reliability. However, compared with the extensive literature on fault detection methods, the systematic setting of accurate and responsive detection threshold has received relatively little attention. This article provides an analytical investigation of the most existing fault detection based-threshold setting. After that, a comparative study of global through exponentially weighted moving average and local using peak-to-peak method thresholding strategies was conducted using a laboratory-scale conveyor-based sorting system under two spontaneous faults scenarios. The results demonstrated that while each threshold method provides advantages, their efficiency is highly dependent on fault dynamics and operational conditions. Finally, the outline of future research direction towards a unified adaptive threshold framework that combines the strength of global and local approaches and supports reliable deployment in industrial environments.