Induction Motor Stator Winding Inter-Tern Short Circuit Fault Detection Based on Start-Up Current Envelope Energy

Liting Chen, Jianhao Shen, Gang Xu, Cheng Chi, Qiaohui Feng, Yang Zhou, Yuanzhi Deng, Huajie Wen2023 · DOI

TL;DR

This paper introduces a novel method for detecting early-stage inter-turn short circuit (ITSC) faults in induction motors by analyzing the energy of the start-up current envelope. The approach achieves high accuracy in identifying faults and their severity, offering a more robust solution than traditional methods, especially under challenging conditions like no-load operation.

What Problem Does This Paper Address?

Induction motors are workhorses in industry, but they are prone to faults. Inter-turn short circuits (ITSC) in the stator winding are a common electrical fault, accounting for a significant portion of motor failures. Detecting these faults early is crucial to prevent catastrophic damage and costly downtime. However, early-stage ITSC faults produce subtle electrical signatures, making them difficult to detect, especially when masked by noise, unstable power, or during the high-current transient of motor start-up. Existing methods often rely on steady-state operation, which may not capture these early fault indicators effectively.

Methods

The proposed method tackles this challenge by focusing on the start-up current, a period where fault signatures are amplified. The core steps are:

  • Data Acquisition: Measuring the three-phase current during motor start-up.
  • Feature Extraction:
    • Sliding-Average Filter: Smoothing the raw current signal to reduce noise.
  • Envelope Extraction: Using the Akima interpolation function to derive the upper and lower envelopes of the filtered signal. Akima interpolation is chosen for its robustness and ability to produce smooth curves, unlike cubic spline interpolation which can be sensitive to outliers and cause oscillations.
  • Gaussian Window Weighting: Applying a Gaussian window to the extracted envelopes to mitigate "end effects" caused by variations in the initial phase angle of the voltage source. This step amplifies the amplitude variations at the start of the current signal.
  • Envelope Energy Calculation: Computing the energy of the weighted upper and lower envelopes. This energy value serves as the fault feature.
  • Classification: Employing a Support Vector Machine (SVM) model to classify the extracted energy features into different health states: healthy, early-stage fault (fault1), and severe fault (fault2). SVM is selected for its effectiveness in small sample classification and its ability to handle non-linear problems.

Datasets

The study utilized two main sources of data:

  • Simulated Data: A mathematical model of ITSC faults in three-phase induction motors was created to generate simulation data. This allowed for controlled experiments across various fault levels (healthy, μ=0.01, μ=0.03), initial phase angles (0° to 300°), and load conditions (0, 0.25 Nm, 0.5 Nm).
  • Experimental Data: A dedicated motor performance testing platform was built. This platform collected actual start-up current signals from three types of induction motors: a healthy motor, an early-stage fault motor (created by manually scraping insulation and welding coils), and a severe fault motor (created by welding six adjacent coils). Data was collected under no-load, 0.52 Nm, and 1.04 Nm load conditions.

Key Insights and Ideas

  • Start-up Current as a Rich Feature Source: The paper highlights that the transient start-up phase, with its significantly higher current, is a more sensitive period for detecting early ITSC faults than steady-state operation.
  • Akima Interpolation for Robust Envelope Extraction: The choice of Akima interpolation over other methods like Hilbert transform or wavelet transform is justified by its robustness to outliers and its ability to produce smooth, accurate envelopes that pass through extreme points.
  • Gaussian Window for End-Effect Mitigation: The application of a Gaussian window is a clever way to address the common problem of end effects in signal processing, particularly relevant here due to the variable initial phase angles of the AC supply.
  • Envelope Energy as a Discriminative Feature: The energy of the envelope, rather than just the envelope shape itself, proves to be a powerful and discriminative feature for classifying fault severity.
  • SVM's Superiority for Early Faults: The study demonstrates that SVM, when combined with the proposed envelope energy features, outperforms other common machine learning classifiers (LR, KNN, NN) in detecting early-stage ITSC faults, especially under no-load conditions.

Key Results

  • High Overall Accuracy: The proposed method achieved an average detection accuracy of 96.9% for healthy, early-stage, and severe faults across different load conditions.
  • Excellent No-Load Performance: For early-stage fault detection under no-load conditions, the method achieved an impressive average accuracy of 98.8% with SVM, outperforming LR, KNN, and NN.
  • Fault Severity Identification: The method not only detects the presence of ITSC faults but also accurately determines their severity (early-stage vs. severe).
  • 100% Training and Validation Accuracy: In specific experimental scenarios (e.g., early-stage fault detection under no-load), the model achieved 100% accuracy in both training and validation.

Limitations and Future Work

The authors acknowledge two primary limitations:

  • Voltage Source Instability: Fluctuations in the voltage source can impact the start-up current, potentially reducing fault detection accuracy. Future work should consider voltage stability characteristics.
  • Fault Location Ambiguity: The current method, using single-phase current, can detect ITSC faults but cannot pinpoint their exact location within the winding or across phases.

Future research directions include:

  • Multi-phase Current Sensing: Incorporating current sensors on other phases to enable fault location determination.
  • Voltage Signal Integration: Utilizing voltage signals to mitigate the impact of voltage instability.
  • Temperature Sensor Integration: Adding temperature sensors to monitor motor health and analyze the influence of thermal factors.
  • Advanced Machine Learning Models: Exploring more sophisticated machine learning classification models to further enhance fault detection accuracy.

Summary updated

BibTeX
@misc{chen2023induction,
  author = {Liting Chen and Jianhao Shen and Gang Xu and Cheng Chi and Qiaohui Feng and Yang Zhou and Yuanzhi Deng and Huajie Wen},
  title = {{Induction Motor Stator Winding Inter-Tern Short Circuit Fault Detection Based on Start-Up Current Envelope Energy}},
  year = {2023},
  doi = {10.3390/s23208581},
}
Figure