Piecewise Negative Sequence Voltage Estimation for Stator Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor-less techniques for detecting stator winding faults in electrical machines face accuracy issues due to nonlinear dependencies on load and supply voltage, leading to poor detection of inter-turn faults, as linear estimation techniques are inadequate and computationally intensive methods are costly and difficult to implement.
Innovation Solution
A diagnostic system using a processor to compute positive, negative, and zero sequence components from three-phase voltages and currents, employing a modified recursive least square (RLS) method for a two-step initialization algorithm to accurately estimate noise factor contributions and isolate stator fault contributions to the negative sequence voltage, thereby improving fault detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If linear estimation techniques (LMS, RLS) are used to estimate noise factor contribution, then the implementation is simple and computationally efficient, but the accuracy of noise factor contribution estimation deteriorates due to nonlinear dependencies on load and supply voltage
Solution Approach 1:
The patent segments the estimation process into two distinct phases: an initialization phase that computes lookup tables (LUTs) offline using linear techniques, and an operational phase that uses piecewise linear approximation with pre-computed coefficients. This segmentation allows complex nonlinear estimation to be broken down into manageable linear segments that can be efficiently implemented in real-time.
Solution Approach 2:
The patent performs preliminary computations during system initialization and offline processing to create lookup tables containing pre-computed coefficients and noise factor contributions. These pre-computed values are stored for later use during fault detection, eliminating the need for complex real-time nonlinear calculations and enabling accurate yet computationally efficient operation.
2Measurement precision
If complex nonlinear equations, neural networks, or load bin approaches are used to accurately estimate noise factor contribution, then the accuracy of fault detection improves, but the computational complexity and implementation cost increase
Solution Approach 1:
The patent performs preliminary computations during system initialization and offline processing to create lookup tables containing pre-computed coefficients and noise factor contributions. These pre-computed values are stored for later use during fault detection, eliminating the need for complex real-time nonlinear calculations and enabling accurate yet computationally efficient operation.
Solution Approach 2:
The patent creates simplified copies of the complex nonlinear relationships by generating lookup tables that approximate the nonlinear behavior using piecewise linear models. These tabulated representations capture the essential nonlinear characteristics without requiring complex computational resources during operation, effectively copying the behavior of complex models in a simplified form.
3Productivity
If linear optimization techniques are used for noise factor estimation, then the computational load is reduced and implementation is simpler, but the lowest severity of inter-turn fault that can be detected increases (worsens)
Solution Approach 1:
The patent segments the estimation process into an offline initialization phase where lookup tables are computed, and an online operational phase where piecewise linear approximation is applied. This allows the system to achieve both computational efficiency in real-time operation and high accuracy in fault detection by leveraging pre-computed segment-specific coefficients.
Solution Approach 2:
The patent changes the parameters used in estimation by switching from fixed linear coefficients to piecewise linear coefficients that vary based on operating conditions (load and voltage). By adapting the estimation parameters to match current operating conditions using the lookup tables, the system maintains high sensitivity for detecting low-severity faults while preserving computational efficiency.
Data Source
AI summary
A diagnostic system configured to detect a stator winding fault in an electrical machine comprising a plurality of stator windings is provided. The diagnostic system includes a processor programmed to receive measurements of three-phase voltages and currents provided to the electrical machine, compute positive, negative, and zero sequence components of voltage and current from the three-phase voltages and currents, and identify a noise factor contribution and a stator fault contribution to the negative sequence voltage by performing a two-step initialization algorithm comprising a modified recursive least square (RLS) method, the noise factor contribution comprising unbalance in the electrical machine resulting from one or more of positive sequence current, negative sequence current, and positive sequence voltage. The processor is still further programmed to detect a stator fault in the electrical machine based on the stator fault contribution to the negative sequence voltage.