TL;DR
This paper introduces a novel fault diagnosis method using single-valued neutrosophic sets (SVNS) and an improved weighted correlation coefficient. By effectively handling uncertain and imprecise data, the proposed method significantly enhances diagnostic accuracy for mechanical equipment, outperforming existing approaches.
What Problem Does This Paper Address?
As mechanical equipment becomes increasingly automated, the complexity of failure causes and their interdependencies also rise. This leads to uncertain and inaccurate information during fault diagnosis, making traditional deterministic methods insufficient. Existing approaches using neutrosophic sets have limitations in comprehensively considering the relationships between different fault types and ideal scenarios, potentially leading to misdiagnoses. This paper aims to address this by developing a more robust fault diagnosis framework that can better manage uncertainty and improve diagnostic accuracy.
Methods
The core of this paper lies in the development of a Single-Valued Neutrosophic Set (SVNS) algorithm for fault diagnosis. The process begins by representing fault data and test samples as triangular fuzzy numbers. These fuzzy numbers are then used to generate single-valued neutrosophic sets, which capture truth-membership, indeterminacy-membership, and falsity-membership degrees.
A key innovation is the introduction of an improved weighted correlation coefficient. This coefficient is designed to more comprehensively analyze the relationship between the SVNS of the fault data and an ideal SVNS representing fault templates. Unlike previous methods that might only consider maximum values, this improved coefficient accounts for the overall similarity and dissimilarity across all membership degrees. The algorithm then ranks these correlation coefficients to determine the most likely fault.
The detailed steps involve:
- Collecting data and generating triangular fuzzy numbers.
- Calculating membership degrees and forming single neutrosophic sets.
- Constructing a single-valued neutrosophic set decision matrix.
- Generating an ideal single-valued neutrosophic number for each attribute.
- Calculating the improved weighted correlation coefficient between the decision matrix and the ideal neutrosophic number.
- Sorting these coefficients to identify the fault.
Datasets
The paper uses data from a motor rotor fault diagnosis example as described in reference [11]. This data was analyzed using LABVIEW software. For validation, the authors extracted arbitrary 40 faults data from three fault templates, totaling 120 fault samples. These were used to compare the proposed algorithm's accuracy against a baseline SVNPWA algorithm.
Key Insights and Ideas
- Neutrosophic Sets for Uncertainty: The paper highlights the power of neutrosophic sets in handling the inherent uncertainty, imprecision, and inconsistency found in real-world fault diagnosis data, which is often a significant challenge.
- Improved Correlation for Better Discrimination: The novel weighted correlation coefficient is crucial. By considering all three membership degrees (truth, indeterminacy, falsity) and incorporating weights, it provides a more nuanced measure of similarity, leading to better differentiation between various fault types.
- Comprehensive Fault Analysis: The method's ability to consider both the single-valued neutrosophic set data and an ideal neutrosophic set data for various faults allows for a more holistic and accurate diagnostic process.
- Practical Relevance: The proposed method is shown to significantly improve diagnostic accuracy, making it more suitable for practical applications where reliable fault identification is critical for safety and economic efficiency.
Key Results
The proposed ISVNS algorithm demonstrated a fault diagnosis accuracy of 99.16% on a test set of 120 fault samples. This represents an improvement of 0.83% compared to the SVNPWA algorithm, which achieved 98.33% accuracy. The illustrative example of the motor rotor fault diagnosis also showed that the analyzed samples belonged to the X fault type with a high correlation coefficient (0.8126).
Limitations and Future Work
The authors acknowledge that the data collection and aggregation for technical faults is a laborious process. Furthermore, the parameters in the diagnosis process were automatically generated in a laboratory simulation, without direct human intervention.
For future work, the authors propose focusing on automating the proposed method to facilitate its practical application and scalability. This would involve developing more efficient data acquisition and parameter tuning strategies.
Summary updated