Enhanced probabilistic neural network with data imputation capabilities for machine-fault classification

Abstract

This paper presents the expectation-maximization (EM) variant of probabilistic neural network (PNN) as a step toward creating an autonomous and deterministic PNN. In the real world, faulty reading sensors can happen and will create input vectors with missing features yet they should not be discarded. To overcome this, regularized EM is put in place as a preprocessing step to impute the missing values. The problem faced by users when using random initialization is that they have to define the number of clusters through trial and error, which makes it stochastic in nature. Global k-means is used to autonomously find the number of clusters using a selection criterion and deterministically provide the number of clusters needed to train the model. In addition, fast Global k-means will be tested as an alternative to Global k-means to help reduce computational time. Tests are conducted on both homoscedastic and heteroscedastic PNNs. Benchmark medical datasets and also vibration data collected from a US Navy CH-46E helicopter aft gearbox known as Westland were used. The tests' results fully support the usage of fast Global k-means and regularized EM as preprocessing steps to aid the EMtrained PNN.

Revisiting probabilistic neural networks: a comparative study with support vector machines and the microhabitat suitability for the Eastern Iberian chub (Squalius valentinus)

“…Although homoscedastic PNNs demonstrated great performance, a single global smoothing parameter may be insufficient to achieve the desired accuracy (Chang et al, 2009). By adapting separate smoothing parameters for each coordinate or variable (i.e., 1 ≠ 2 ≠ ⋯ ≠ ), the classification accuracy can be greatly improved (Specht and Romsdahl, 1994), as has been corroborated by a number of studies (e.g., Kusy and Zajdel, 2015;Li and Ma, 2008).…”

“…Currently, a number of different clustering algorithms have been used as pre-treatments prior to the development of cluster PNNs, such as global k-means (Chang et al, 2009) or j-means (Li and Ma, 2008) ( Fig. 4 -Lower left panel).…”

“…PNNs have been proven to be proficient in various tasks, such as: risk assessment (Adeli and Panakkat, 2009), bacterial growth prediction (Hajmeer and Basheer, 2002), fault detection (Chang et al, 2009) or cancer diagnosis (Berrar et al, 2003). Conversely, to the best of our knowledge, there are very few examples of their use in ecology, despite their having demonstrated great performance (Muñoz-Mas et al, 2014;Siira et al, 2009;Corne et al, 2004) and stability over various prevalence datasets (Muñoz-Mas et al, 2014).…”

The Morlet and Daubechies Wavelet Transforms for Fatigue Strain Signal Analysis

“…However, it has been found that the WT is a new approach in fatigue signal analysis. The idea of this study came from assessment of current research trends, where during the last decade, the transformation method has been used extensively in some previous vibrational random signal analysis research attempts [8][9][10][11]. The fatigue strain signal type was hypothesized to be similar to the vibrational random signal pattern.…”