On Designing Features for Condition Monitoring of Rotating Machines

TL;DR

Problem: feature design for CBM of rotating machines is largely empirical and data-dependent. Method: a histogram-theory–driven end-to-end feature extraction framework uses state-specific bin width w(X) = 3.49 * sigma / N^(1/3) to design features, producing vectors with length m_i = (X^(i)_max - X^(i)_min) / w(X^(i)). Contributions: introduces this principled design and validates it with NN, RF, and SVM classifiers on acoustic, CWRU, and IMS datasets, achieving near-perfect accuracies. Significance: simplifies CBM pipelines by unifying feature design across sensor modalities for real-time machine health-state recognition.

Abstract

Various methods for designing input features have been proposed for fault recognition in rotating machines using one-dimensional raw sensor data. The available methods are complex, rely on empirical approaches, and may differ depending on the condition monitoring data used. Therefore, this article proposes a novel algorithm to design input features that unifies the feature extraction process for different time-series sensor data. This new insight for designing/extracting input features is obtained through the lens of histogram theory. The proposed algorithm extracts discriminative input features, which are suitable for a simple classifier to deep neural network-based classifiers. The designed input features are given as input to the classifier with end-to-end training in a single framework for machine conditions recognition. The proposed scheme has been validated through three real-time datasets: a) acoustic dataset, b) CWRU vibration dataset, and c) IMS vibration dataset. The real-time results and comparative study show the effectiveness of the proposed scheme for the prediction of the machine's health states.

Figures (6)

• Figure 1: Illustration of constructing histogram for a dataset. $(a)$ Dataset: $\textbf{X}= \{x_1, x_2, \cdots, x_N\}$ . $(b)$ Histogram, where $w(\textbf{X})=t_{k+1}-t_k$ denotes bin width having bin interval $[t_k, t_{k+1})$
• Figure 2: The proposed end-to-end framework for condition based monitoring of rotating machines. In this framework, we leverage the histogram theory to determine the appropriate bin width for each state, which is then used to design input features for the corresponding state. These input features are then fed into various classifiers to predict the health state of machines.
• Figure 3: Illustration of selecting maximum and minimum amplitudes from the dataset. $(a)$ Data for the $i^{th}$ health state of the machine, i.e., $\textbf{X}^{(i)}= \{x_1, x_2, \cdots, x_N\}$. $(b)$ Plot to depict maximum $(\textbf{X}^{(i)}[max])$ and minimum $(\textbf{X}^{(i))}[min]$ amplitudes.
• Figure 4: Visualization of features for acoustic dataset. Left: raw sensor data and Right: Extracted features using the proposed algorithm.
• Figure 5: Visualization of features for vibration dataset. Left: raw sensor data and Right: Extracted features using the proposed algorithm.