This paper presents a method for a variable fault detection algorithm, based on wavelets, and offers the possibility to realize soft, hard and very hard fault detection. The proposed algorithm is based on the estimation of the variance of the local Lipschitz constant of the signal over a receding time horizon. The fault (outlier) is recognized if the local Lipschitz constant lies outside the computed boundary. Currently, to estimate parameters in the nano range. Thus the input signal requires a very high frequency, which subsequently requires a very high sampling rate. A modified Recursive Least Squares (RLS) method was used to estimate parameters of conductive multifilament fibers using on-line identification model during the manufacturing process, including inductance, within the nano range, using input-output scaling factors. In contrast, this technique uses a broader sampling rate and an input signal with low frequency to identify the parameters characterizing the linear model. This method was used to provide a scaled identification bandwidth together with a reduced sampling rate. Through the scaling of the input-output data, a general model of the identification technique was obtained to estimate time-varying sinusoidal disturbance signal during the manufacturing process. In our contribution a time-varying sinusoidal disturbance is considered in terms of magnitude and frequency. The identification is obtained using modified Least Squares Methods (LSM). In this kind of system, the inductance represents the most critical parameter to be estimated. The examined results indicated that the proposed RLS algorithm method, using a forgetting factor, is a useful method for estimating time-varying sinusoidal disturbances as well as the inductance.