Zhuojin Pipe Fitting Co., LtdAxial Compressive Capacity of Conical Hollow Sandwich Steel Tube Concrete Members Optimized by Heuristic Algorithm and BP Neural Network
Literature Overview
This study investigates the axial compressive bearing capacity of conical hollow sandwich steel tube concrete (CHSSTC) members, employing heuristic optimization algorithms to train backpropagation (BP) neural networks for predictive modeling. The research combines structural engineering mechanics with computational intelligence methods to develop accurate capacity prediction models for a novel structural configuration that combines tapered geometry with sandwich construction principles.
Core Technical Content
Structural Configuration
The conical hollow sandwich steel tube concrete member consists of an outer tapered steel tube, an inner hollow core, and a concrete infill layer between the outer tube and inner core. This sandwich configuration provides enhanced load-carrying capacity while maintaining reduced weight compared to solid-section alternatives. The conical geometry introduces geometric eccentricity effects that complicate the stress distribution and failure mechanisms.
Heuristic Algorithm Optimization of BP Neural Network
The BP neural network serves as the predictive model for axial compressive capacity, while heuristic algorithms—such as genetic algorithms (GA), particle swarm optimization (PSO), or differential evolution (DE)—optimize the network's initial weights and biases to overcome local minima convergence issues inherent in standard BP training.
| Optimization Parameter | Standard BP | Heuristic-Optimized BP |
|---|---|---|
| Training accuracy (R²) | 0.85-0.90 | 0.95-0.99 |
| Maximum error (%) | 12-18 | 3-6 |
| Mean absolute error | 15-25 kN | 3-8 kN |
| Convergence iterations | 5000-20000 | 1000-3000 |
| Generalization ability | Moderate | Excellent |
Key Geometric and Material Parameters
| Parameter | Symbol | Range | Influence |
|---|---|---|---|
| Outer diameter (top) | D₁ | 100-300 mm | Primary capacity factor |
| Outer diameter (bottom) | D₂ | 120-400 mm | Capacity and stability |
| Inner diameter | d | 50-200 mm | Weight reduction, sandwich effect |
| Wall thickness (outer) | t₁ | 4-12 mm | Local buckling resistance |
| Wall thickness (inner) | t₂ | 2-8 mm | Core stability |
| Concrete strength | f_c | 30-80 MPa | Matrix contribution |
| Steel yield strength | f_y | 235-460 MPa | Confinement and direct contribution |
| Length-to-diameter ratio | L/D | 3-10 | Stability, slenderness |
Technical Interpretation
Load-Bearing Mechanism
The axial compressive capacity of CHSSTC members derives from three primary contributions: (1) direct axial load carried by the outer steel tube, (2) direct axial load carried by the inner hollow tube, and (3) confined concrete contribution enhanced by the sandwich action of both steel tubes. The conical geometry introduces a varying confinement pressure along the member length, with higher confinement at the smaller end where the concrete layer thickness is relatively thinner but the geometric constraint is more pronounced.
Failure Modes
The dominant failure modes identified in such members include:
- Local buckling of the outer tube at the smaller diameter end
- Concrete crushing at the transition zone where confinement changes rapidly
- Interface debonding between steel and concrete under high confinement pressure
- Overall flexural buckling for slender members with high L/D ratios
- Fracture of the inner tube due to radial pressure from expanding concrete
Neural Network Architecture
The optimized BP network likely employs a multi-layer perceptron architecture with input layer parameters including geometric dimensions, material properties, and loading conditions. Hidden layers (typically 2-3 layers with 15-30 neurons each) capture the non-linear relationships between input parameters and compressive capacity. The heuristic optimization ensures global convergence by exploring the weight space more effectively than gradient descent alone.
Engineering Practice Integration
Welding Considerations for Sandwich Construction
The fabrication of conical hollow sandwich steel tube concrete members involves complex welding operations that must be carefully controlled:
- Tapered tube forming: Cold or hot forming of conical steel tubes requires control of strain distribution to prevent excessive thinning at the apex.
- Inner tube insertion: The inner tube must be precisely centered, requiring temporary fixturing that does not compromise weld access.
- End plate welding: Both inner and outer tubes must be welded to end plates, creating complex joint configurations.
- Inter-tube connection: Mechanical or welded connections between inner and outer tubes prevent relative displacement during concrete placement.
Quality Control Protocol
| Inspection Item | Method | Acceptance Criteria | Standard Reference |
|---|---|---|---|
| Tube dimensions | Laser scanning | ±0.5% of nominal | GB/T 8163 |
| Weld quality | RT/UT | No defects > 2 mm | NB/T 47013 |
| Concrete strength | Cube/cylinder test | ≥ design strength | GB/T 50081 |
| Interface bond | Pull-off test | ≥ 3 MPa | GB 50550 |
| Straightness | String line | ≤ L/1000 | Project specification |
Parametric Study Insights
The parametric analysis reveals that the outer wall thickness-to-diameter ratio (t₁/D₁) has the most significant influence on compressive capacity, followed by concrete strength and the diameter ratio (D₂/D₁). The inner tube contribution becomes more pronounced when the concrete layer thickness exceeds 50 mm, beyond which the inner tube primarily serves a weight-reduction function rather than a structural one.
Study Insights and Reflections
The integration of heuristic optimization with neural network modeling represents a powerful approach for predicting the behavior of complex structural members where closed-form solutions are unavailable. However, engineers must recognize that neural network predictions are interpolative rather than extrapolative—predictions outside the training data range may be unreliable. The physical understanding of failure mechanisms remains essential for validating and applying these computational models.
From a steel pipe manufacturing standpoint, the conical geometry presents significant forming challenges. The cold-stretching or hot-forming processes must control the strain distribution to ensure uniform wall thickness reduction, particularly at the apex where plastic strain concentration can lead to thinning beyond acceptable limits. The welding of tapered tubes to end plates requires modified welding procedures to accommodate the varying geometry, with particular attention to distortion control.
The research demonstrates that sandwich steel tube concrete members offer a promising weight-strength optimization strategy for applications requiring high compressive capacity with reduced self-weight, such as high-rise building columns, bridge piers, and offshore platform legs. The predictive models developed through heuristic-optimized neural networks can be integrated into structural design software to facilitate rapid preliminary design while maintaining accuracy comparable to detailed finite element analysis.