TL;DR
This paper introduces an intelligent, real-time method using a Back Propagation Neural Network (BPNN) to analyze the correlated mechanical properties of members during the hoisting of large-span valve hall grid structures. This approach aims to overcome the limitations of traditional single-point analysis by providing a more comprehensive understanding of structural integrity and enabling early detection of potential risks.
What Problem Does This Paper Address?
The construction and hoisting of super-span valve hall grid structures are complex engineering feats where structural safety and quality are paramount. Traditional methods for monitoring the mechanical properties of these structures rely on manual analysis of sensor data and expert experience. While these methods can provide some insights, they struggle to capture the intricate correlations between different members and offer a holistic view of the entire structure's safety during the dynamic hoisting process. This often leads to challenges in accurately assessing overall structural integrity, identifying hidden dangers, and pinpointing specific members at risk, especially given the complex environments and potential for signal interference during construction.
Methods
The core of this paper is an intelligent correlation real-time analysis method built upon a Back Propagation Neural Network (BPNN). The approach focuses on modeling the mechanical properties of a target measuring point by considering the correlated data from multiple other measuring points within the grid structure.
Here's a breakdown of the methodology:
- Multiple-Measuring-Point Correlation Model: Instead of analyzing each sensor's data in isolation, the method establishes a model where the mechanical data from n-1 monitoring points are used as input variables to predict the mechanical data of the nth monitoring point. This is iteratively applied to all points, creating a comprehensive understanding of interdependencies.
- BPNN Implementation: A three-layer BPNN is employed for learning and prediction. The input layer receives the mechanical features from multiple points, the hidden layer processes these features, and the output layer predicts the mechanical state of a target point.
- Intelligent Real-Time Analysis System: The developed correlation model is integrated into an intelligent system that manages and applies the mechanical property correlations and identifies abnormalities in real-time. This system allows for visualization and analysis of the 3D model, enabling users to inspect the state of specific nodes or members.
- Abnormal Member Identification: The system identifies abnormal members by comparing the predicted mechanical properties with actual on-site monitoring data. It leverages the BPNN's ability to better solve contribution distribution problems to understand how different members influence a reference point (E030 in their study).
Datasets
The study utilized data collected from a super-span valve hall grid structure hoisting project in South China.
- Data Collection: Stress observations were carried out at four stages of member installation and hoisting.
- Monitoring Points: A total of 30 monitoring points (E001–E030) were arranged along the axis of the valve hall. One point (E004) was eliminated due to damage.
- Training and Testing: The data from 29 monitoring points were used as input to predict the behavior of the remaining point, with 90% of the sample data used for training and 10% for testing. The study also used numerical simulation data alongside on-site monitoring data for abnormal member identification.
Key Insights and Ideas
- Shifting from Single-Point to Correlated Analysis: The most significant insight is the move away from analyzing individual sensor readings to understanding the complex interplay between different structural members. This is crucial for large, interconnected structures like super-span valve halls.
- BPNN as a Powerful Tool for Correlation: The paper demonstrates the effectiveness of BPNNs in capturing and modeling these complex, non-linear correlations, which are difficult to discern through traditional methods.
- Real-time Monitoring and Early Warning: The development of an intelligent real-time analysis system is key. It transforms raw data into actionable insights, allowing for immediate identification of potential issues and proactive intervention.
- Predictive Power for Hidden Dangers: By modeling correlations, the system can predict how changes in one part of the structure might affect others, effectively "seeing" potential problems before they become critical.
Key Results
- High Prediction Accuracy: The developed multiple-measuring-point mechanical property correlation model achieved good results, with the coefficient of determination (R²) ranging from 0.85 to 0.98 across various time series and members. This indicates a high degree of accuracy in predicting mechanical properties.
- Effective Identification of Abnormal Members: The system successfully identified members with potential risks. For instance, members 17 and 20 showed stresses exceeding the first-level limit, and members 6 and 7 were identified as being at lifting points that experienced disturbances.
- Qualitative Safety Evaluation: The "degree of confusion" metric was introduced to qualitatively evaluate the overall safety pattern, showing a small increase (6.38%) in confusion between numerical simulation and on-site monitoring, suggesting the structure remained in a relatively safe state.
Limitations and Future Work
- Limitations:
- The study acknowledges that while some members exceeded the first-level limit, they did not reach the third-level limit, indicating they were still within a safe state. However, the paper doesn't extensively discuss the implications of exceeding lower-level limits or the precise thresholds for "danger."
- The numerical simulation considered only three stages of hoisting, which might not fully capture all dynamic aspects.
- The "degree of confusion" metric is a novel approach, and its broader applicability and validation across different structures might require further research.
- Future Work:
- The paper suggests that the established model provides a new reference for the normal operation and maintenance of super-span valve hall grids, implying its potential for long-term structural health monitoring.
- Further research could explore the application of this method to other large-span structures and investigate more advanced correlation analysis techniques.
- Expanding the numerical simulation to encompass more dynamic hoisting phases could enhance the model's robustness.
Summary updated