TL;DR
This paper introduces a novel fault diagnosis method that leverages multi-stage fault template data and neutrosophic sets to effectively handle the inherent uncertainty and imprecision in real-world systems. By integrating data from various stages and employing a simplified neutrosophic weighted averaging operator, the method achieves more robust and accurate fault identification compared to traditional approaches.
What Problem Does This Paper Address?
Fault diagnosis is crucial for maintaining the reliability of systems across various domains, from manufacturing to aerospace. However, a significant challenge in this field is the inherent uncertainty, imprecision, and incompleteness of fault information. This uncertainty arises from the complex and dynamic nature of actual operating environments, leading to variations in collected data even for the same fault. Furthermore, different faults can exhibit similar features, and the same feature might be a symptom of multiple faults, complicating diagnosis. Traditional methods often struggle to adequately address these issues, particularly when dealing with multi-stage data that reflects temporal variations. This paper aims to bridge this gap by proposing a method that can effectively manage these uncertainties using multi-stage fault template data and the mathematical framework of neutrosophic sets.
Methods
The core of the proposed method lies in the application of neutrosophic sets to represent and fuse uncertain fault information. The process involves several key steps:
- Multi-Stage Data Collection: Data for known fault types and an unknown fault sample is collected across multiple stages (e.g., different time intervals or operating conditions). This captures the temporal dynamics of fault signatures.
- Normal Distribution Modeling: For each stage and feature, a normal distribution model is established for both known fault types and the unknown sample. This is done by calculating the arithmetic average and variance of the collected data.
- Neutrosophic Set (SNS) Generation: Single-valued neutrosophic sets (SVNSs) are generated for each fault type under each feature. The membership degree (truth, indeterminacy, falsity) is determined by comparing the normal distribution of the unknown sample with the normal distributions of known fault types. Specifically, the membership degree is defined as the ratio of the vertical coordinate of the intersection point between the unknown sample's distribution and the fault type's distribution to the peak value of the unknown sample's distribution. The indeterminacy and non-membership degrees are then derived using Shannon entropy and a relationship with the membership degree.
- Aggregation of SNSs:
- Within Features: For each fault type and under each feature, the SVNSs generated from the multiple stages are fused using the Simplified Neutrosophic Weighted Averaging (SNWA) operator. This step aggregates the temporal information for a specific feature.
- Across Features: The fused SNSs for each fault type, now representing aggregated information across all features, are further fused using the SNWA operator. This step consolidates the information from all features for each fault type.
- Defuzzification and Diagnosis: The final aggregated neutrosophic sets for each fault type are defuzzified into crisp numbers. These crisp numbers represent the degree to which the unknown sample supports each fault type. The fault type with the highest crisp number is then identified as the diagnosed fault.
The novelty lies in the integration of multi-stage data with neutrosophic sets and the specific application of the SNWA operator for aggregation, allowing for a more comprehensive handling of temporal uncertainties.
Datasets
The paper uses experimental data from a multi-functional flexible rotor test-bed. This data was collected for three known fault types (rotor imbalance, rotor misalignment, and support base loosening) and an unknown fault sample. For each fault type and feature, data from five stages was collected, with each stage containing forty consecutive observation values over a 16-second interval. The paper references a previous study [57] for the origin of this data.
Key Insights and Ideas
- Multi-stage data is crucial for capturing temporal uncertainty: The paper highlights that data collected at different stages reflects the dynamic nature of fault signatures and environmental variations, which traditional single-stage methods overlook.
- Neutrosophic sets are well-suited for uncertain information: The framework of neutrosophic sets, with its ability to represent truth, indeterminacy, and falsity, effectively models the inherent uncertainty in fault diagnosis.
- SNWA operator provides robust aggregation: The SNWA operator is used to fuse information from multiple stages and multiple features, effectively combining diverse and potentially conflicting data into a coherent representation.
- Integration of multi-stage data and neutrosophic sets overcomes limitations: The proposed method demonstrates that combining these two approaches can lead to more accurate and reliable fault diagnosis, especially when dealing with complex relationships between faults and features.
- Defuzzification provides a clear diagnostic outcome: Converting the aggregated neutrosophic information into crisp numbers allows for a straightforward ranking and identification of the most likely fault type.
Key Results
The proposed method was tested on an illustrative example of a motor rotor fault diagnosis. The results showed that the method correctly identified the fault type for the unknown sample (F1: Rotor imbalance). Furthermore, when applied to two other unknown fault samples (S2 and S3), the method also achieved correct diagnoses, consistent with the results obtained by Xu's method [54]. The paper presents a comparison table (Table 7) showing that both methods achieved identical diagnosis results and rankings for all three samples, underscoring the effectiveness of the proposed approach. The authors also note that their method is more adept at handling the uncertainty of multi-stage fault template data compared to methods that rely on limited stage data.
Limitations and Future Work
The authors acknowledge that while their method effectively handles uncertainty, there is room for improvement.
- Feature Weight Calculation: The paper assumes equal weights for features in the aggregation process. Future work could focus on developing an appropriate method for calculating feature weights based on the collected information.
- Simplification of Aggregation: For convenience in calculation, the authors suggest exploring ways to simplify the double aggregation of neutrosophic sets in future research.
The paper concludes that the proposed method is efficient and convenient for fault diagnosis, particularly in scenarios with complex relationships between fault types and features, and it effectively handles uncertain information.
Summary updated