PSI - Issue 17

Andra Gabriela Stancu et al. / Procedia Structural Integrity 17 (2019) 238–245 A. G. Stancu et al./ Structural Integrity Procedia 00 (2019) 000 – 000

240

3

suited for long-term in-situ maintenance schemes, compared to one-time inspections. With the concept and benefits of condition monitoring introduced, there are still challenges remain to be addressed. To begin with, modularization and extraction of the signs of degradation, i.e. parameters, are essential to establish a normal or healthy condition as a baseline. Moreover, to mitigate the fluctuation in the parameters induced by change of environmental condition, optimal number of recordings should be selected for signal conditioning. Finally, the interpretation of the parameters should be discussed. This paper addresses the aforementioned challenges using a combination of signature extraction, baseline convergence and similarity analysis, which will be elaborated in the following sections. The continuous monitoring technology described in this paper aims to generate warnings at the early stages of damage, allowing enough time for the operators to develop and apply predictive maintenance measures. The methodology used is based on a pattern recognition algorithm for machinery operating under healthy conditions, generating a baseline for the identification of arising deviations from the normal operation. The main objective is to create an alarm when the signals captured from the running process deviate from their mean or when the signal variability increases. The condition monitoring process comprises mainly of three components, namely: Signal Acquisition, Baseline Analysis and Similarity Analysis. The core concept is to utilize raw signals captured from different sensory technologies that are applicable, to develop a system signature, defined as a collection of descriptive signal patterns and relevant parameters of the signal. Once the signature is established, the baseline analysis will be conducted to statistically determine, how many signature states can be collectively taken into consideration as a stable system condition, i.e., a baseline. A sample of data can be regarded as a baseline only when it gathers sufficient information about the mechanism’s running state and being stable enough (i.e. its standard deviation is small) for allowing subtle changes to be detected. Finally, from the establishment of the baseline condition onward, the condition monitoring process will be performed using similarity analysis to assess the deviations of the current condition from the baseline. This process is continuously performed for all subsequent conditions, which by nature should have the same size (i.e. number of states) as the established baseline condition. As described in the previous section, although it is commonly recognized that machinery in operation constantly generates various kinds of system information, such as vibration, noise, increase of temperature, etc., this raw data is either unnoticeable or incomprehensible directly to operators. Sensory technologies are applied to capture such system information, which will then be digitalized and processed using DSP methods and translated into meaningful indications and intuitive trends. The first step of the translation process is to acquire a set of parameters from raw signals to establish the state of the monitored system. Assuming that a series of raw signals with predetermined length L has been captured, and processed to acquire N types of parameters {P i | i = 1, 2…N} for the n th time, then the current (n th ) state of the system can be described as a signature set S n : = [ (1) The set S contains the system ’ s historical state (S 1 to S n-1 ), and the current state S n . The value of recording length, L, for each series of signals is determined to cover a span of at least one period of the lowest rotating speed of the machinery system. The selection of the parameters N, should be investigated on a case-by-case basis. 2.1. System parameters and signature set 1 2 3 ⋮ ] 2. Methodology